The JavaScript AI Tech Stack: SDKs & Frameworks (December 2025)

You've chosen your model. You've estimated your costs. Now you need the tools to actually ship.

The JavaScript AI ecosystem has exploded in 2025. We have production-ready frameworks, type-safe SDKs, and an industry standard for tool integrations. This post covers the first two layers: SDKs and Frameworks.

Layer 1: Provider SDKs

The foundation of any AI feature is the SDK that talks to your model provider.

OpenAI SDK (`openai`)

The most mature option. Excellent TypeScript support, comprehensive documentation, largest community.

import OpenAI from 'openai'
 
const openai = new OpenAI()
 
const response = await openai.chat.completions.create({
  model: 'gpt-5-mini',
  messages: [{ role: 'user', content: 'Hello!' }]
})

import OpenAI from 'openai'
 
const openai = new OpenAI()
 
const response = await openai.chat.completions.create({
  model: 'gpt-5-mini',
  messages: [{ role: 'user', content: 'Hello!' }]
})

Features: Streaming, function calling, batch API, structured outputs, vision.

Anthropic SDK (`@anthropic-ai/sdk`)

Full TypeScript support with tool use and computer use capabilities. Anthropic created MCP—the industry standard for AI-tool integrations.

import Anthropic from '@anthropic-ai/sdk'
 
const anthropic = new Anthropic()
 
const response = await anthropic.messages.create({
  model: 'claude-sonnet-4-5-20251022',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello!' }]
})

import Anthropic from '@anthropic-ai/sdk'
 
const anthropic = new Anthropic()
 
const response = await anthropic.messages.create({
  model: 'claude-sonnet-4-5-20251022',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello!' }]
})

Google AI SDK

Two options:

@google/generative-ai for AI Studio (simpler)
@google-cloud/vertexai for Vertex AI (enterprise)

Google's strength is native multimodal—text, images, video, and audio in a single model.

OpenAI Agents SDK (`@openai/agents`)

New in 2025. A dedicated SDK for building multi-agent workflows.

import { Agent, run } from '@openai/agents'
 
const agent = new Agent({
  name: 'assistant',
  model: 'gpt-5-mini',
  tools: [myTool],
  instructions: 'You are a helpful assistant.'
})
 
const result = await run(agent, 'Help me with my task')

import { Agent, run } from '@openai/agents'
 
const agent = new Agent({
  name: 'assistant',
  model: 'gpt-5-mini',
  tools: [myTool],
  instructions: 'You are a helpful assistant.'
})
 
const result = await run(agent, 'Help me with my task')

Features: Multi-agent handoffs, built-in tracing, guardrails, real-time voice via WebRTC.

When to Use Raw SDKs

Simple, single-provider projects
Maximum control over API calls
Learning and prototyping
When frameworks add unnecessary overhead

Layer 2: Frameworks

Frameworks add structure, provider abstraction, and higher-level patterns.

Vercel AI SDK

2M+ weekly npm downloads. The ecosystem leader for JavaScript AI.

The SDK provides a unified interface across providers—swap OpenAI for Anthropic with one line. Particularly strong for streaming UIs in React and Next.js.

import { generateText } from 'ai'
import { openai } from '@ai-sdk/openai'
 
const result = await generateText({
  model: openai('gpt-5-mini'),
  prompt: 'What is the capital of France?'
})

import { generateText } from 'ai'
import { openai } from '@ai-sdk/openai'
 
const result = await generateText({
  model: openai('gpt-5-mini'),
  prompt: 'What is the capital of France?'
})

AI SDK 6 (Beta as of Dec 2025) added:

Agent abstraction layer (ToolLoopAgent)
Human-in-the-loop tool execution approval
Type-safe agent development
Stable structured output with multi-step tool calling

Note: The agent features are currently in Beta. Core streaming and generation APIs are stable.

Best for: React/Next.js apps, streaming chat UIs, provider flexibility.

LangChain.js + LangGraph

LangChain 1.0 and LangGraph 1.0 shipped in October 2025, bringing full parity with Python.

LangChain is the orchestration layer—chains, prompts, memory, tools. LangGraph adds stateful, multi-agent workflows.

LangGraph's killer features:

Durable execution — Agents persist through failures and resume from last state
Conditional branching — Dynamic routing based on LLM decisions
Human-in-the-loop — Pause for approval, then continue
Checkpointing — Rewind to previous states for debugging
LangGraph Studio — Visual IDE for debugging agent flows
LangGraph Cloud — Managed deployment and scaling

Best for: Complex multi-step agents, enterprise workflows, teams wanting built-in observability via LangSmith.

When to avoid: Simple chat apps (overkill), early learning (too much abstraction).

Mastra

TypeScript-first AI framework, Y Combinator backed, founded in 2024.

Mastra takes an all-in-one approach: agents, workflows, RAG, memory, and evaluations in a single framework.

Key features:

40+ LLM providers through standard interface
Graph-based workflow engine with branching and parallel execution
MCP server support built-in
Built-in observability platform
Local development server with visual playground

Best for: Teams wanting opinionated TypeScript patterns, greenfield projects, developers who prefer batteries-included.

Which Framework Should You Use?

Streaming chat UI? → Vercel AI SDK

Multi-step agents with complex control flow? → LangChain/LangGraph

RAG + agents in one framework? → Mastra or LangChain

Maximum simplicity? → Vercel AI SDK

Built-in observability? → LangChain (LangSmith) or Mastra

TypeScript-first everything? → Mastra

🎁 Free Resource: The Tech Stack Cheat Sheet

Want a one-page reference for this entire stack? I've built a printable cheat sheet with a decision tree and code snippets.

View the AI Tech Stack Cheat Sheet →

What's Next?

You now know the SDK and framework layer of the JavaScript AI stack.

Next post: Building AI Apps: Data, Observability & Deployment — LlamaIndex, vector databases, MCP, observability tools, and deployment platforms.

Previous posts in this series: